Optimality of general lattice transformations with applications to … · 2016-06-24 · lattice...
Transcript of Optimality of general lattice transformations with applications to … · 2016-06-24 · lattice...
Optimality of General Lattice Transformations
with Applications to the Bain Strain in Steel
Konstantinos Koumatos∗
Gran Sasso Science Institute,
Viale Fransesco Crispi 7,
67100 L’Aquila, Italy
Anton Muehlemann†
Mathematical Institute, University of Oxford,
Andrew Wiles Building, Radcliffe Observatory Quarter, Woodstock Road,
Oxford OX2 6GG, United Kingdom
Dated: June 24, 2016
This article provides a rigorous proof of a conjecture by E.C. Bain
in 1924 on the optimality of the so-called “Bain strain” based on a
criterion of least atomic movement. A general framework that explores
several such optimality criteria is introduced and employed to show the
existence of optimal transformations between any two Bravais lattices. A
precise algorithm and a GUI to determine this optimal transformation is
provided. Apart from the Bain conjecture concerning the transformation
from face-centred cubic to body-centred cubic, applications include the
face-centred cubic to body-centred tetragonal transition as well as the
transformation between two triclinic phases of Terephthalic Acid.
MSC (2010): 74N05, 74N10
Keywords: lattice transformation, least atomic movement, Bravais
lattices, Bain strain in steel, fcc-to-bcc, Terephthalic Acid
∗[email protected]†[email protected]
1
arX
iv:1
510.
0470
8v5
[m
ath-
ph]
23
Jun
2016
1 INTRODUCTION
Contents
1 Introduction 2
2 Preliminaries 5
3 Metrics and equivalence on matrices and lattices 9
4 Optimal lattice transformations 11
4.1 The Bain strain in fcc-to-bcc . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Stability of the Bain strain . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3 Terephthalic Acid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5 Concluding remarks 25
1 Introduction
In his seminal article on “The nature of martensite” Bain [Bai24] proposed a mech-
anism that transforms the face-centred cubic (fcc) lattice to the body-centred cubic
(bcc) lattice, a phase transformation most importantly manifested in low carbon
steels. He writes:
“It is reasonable, also, that the atoms themselves will rearrange [...] by a
method that will require least temporary motion. [...] A mode of atomic
shift requiring minimum motion was conceived by the author [...] ”
Bain, 1924
The key observation that led to his famous correspondence was that “If one regards
the centers of faces as corners of a new unit, a body-centered structure is already
at hand; however, it is tetragonal instead of cubic”. He remarks that this is not
surprising “as it is the only easy method of constructing a bcc atomic structure from
the fcc atomic structure”.
Even though now widely accepted, his mechanism, which he illustrated with a
model made of cork balls and needles (see Fig. 1), was not without criticism from
his contemporaries. In their fundamental paper Kurdjumov & Sachs [KS30] wrote
[free translation from German] that “nothing certain about the mechanism of the
martensite transformation is known. Bain imagines that a tetragonal unit cell
within the fcc lattice transforms into a bcc unit cell through compression along one
direction and expansion along the two other. However a proof of this hypothesis
is still missing”.1 Interestingly, without being aware of it, the authors implicitly
used the Bain mechanism in their derivation of the Kurdjumov & Sachs orientation
relationships (see [KM] for details).
1“Uber den Mechanismus dieser ,,Martensitumwandlung” ist bisher nichts Sicheres bekannt.Bain stellt sich vor, daß eine tetra- gonalkorperzentrierte Elementarzelle des Austenits durchSchrumpfung in der einen Richtung und Ausdehnung in den beiden anderen in die kubis-
2
1 INTRODUCTION
Figure 1: From E.C. Bain: the small models show the fcc and bcc unit cells; the largemodels represent 35 atoms in an fcc and bcc arrangement respectively.
In subsequent years, the determination of the transformation mechanism remained
of great interest. In their paper on “Atomic Displacements in the Austenite-
Martensite Transformation“ [JW48] Jaswon and Wheeler again acknowledged that
“Off all the possible distortions of a primitive unit cell of the face-centred
cubic structure, which could generate a body-centred cubic structure of
the given relative orientation, the one which actually occurs is the small-
est” Jaswon and Wheeler, 1948
By combining it with experimental observations of the orientation relationships they
devised an algorithm to derive the strain tensor. However their approach is only
applicable to cases where the orientation relationship is known a-priori.
With the years passing and a number of supporting experimental results (for a
discussion see e.g. [BW72]) the Bain mechanism rose from a conjecture to a widely
accepted fact. Nevertheless, almost a century after Bain first announced his cor-
respondence a rigorous proof based on the assumption of minimal atom movement
has been missing. Of course, the transformation from fcc-to-bcc is not the only
instance where the determination of the transformation strain is of interest. The
overall question remains the same: Which transformation strain(s), out of all the
possible deformations mapping the lattice of the parent phase to the lattice of the
product phase, require(s) the least atomic movement?
To provide a definite answer to this question one first needs to quantify the notion
of least atomic movement in such a way that it does not require additional input
from experiments. Then one needs to establish a framework that singles out the op-
timal transformation among the infinite number of possible lattice transformation
strains. One way to appropriately quantify least atomic movement is the criterion
of smallest principal strains as suggested by Lomer in [Lom55]. In his paper, Lomer
compared 1600 different lattice correspondences for the β to α phase transition in
Uranium and concluded that only one of them involved strains of less than 10%.
chraumzentrierte des α-Eisens ubergeht. Eine Bestatigung fur diese Anschauung konnte bishernicht erbracht werden.”
3
1 INTRODUCTION
More recently, in [CSTJ16] an algorithm is proposed to determine the transforma-
tion strain based on a similar minimality criterion (see Remark 6) that also allows
for the consideration of different sublattices. The present paper considers a cri-
terion of least atomic movement in terms of a family of different strain measures
and, for each such strain measure, rigorously proves the existence of an optimal
lattice transformation between any two Bravais lattices.2 As a main application, it
is shown that the Bain strain is the optimal lattice transformation from fcc-to-bcc
with respect to three of the most commonly used strain measures.
The structure of the paper is as follows: after stating some preliminaries in Sec-
tion 2 we explore in more depth some mathematical aspects of lattices in Section 3.
This section is mainly intended for the mathematically inclined reader and may be
skipped on first reading without inhibiting the understanding of Section 4, which
constitutes the main part of this paper. In this section we establish a geometric
criterion of optimality and prove the existence of optimal lattice transformations for
any displacive phase transition between two Bravais lattices. Additionally a precise
algorithm to compute these optimal strains is provided. In the remaining subsec-
tions, the general theory is applied to prove the optimality of the Bain strain in an
fcc-to-bcc transformation, to show that the Bain strain remains optimal in an fcc
to body-centred tetragonal (bct) transformation and finally to derive the optimal
transformation strain between two triclinic phases of Terephthalic Acid. Similarly
to the fcc-to-bcc transition, this phase transformation is of particular interest as it
involves large stretches and thus the lattice transformation requiring least atomic
movement is not clear.
2In particular, no assumptions are made on the type of lattice points (e.g. atoms, molecules) oron the relation between the point groups of the two lattices.
4
2 PRELIMINARIES
2 Preliminaries
Throughout it is assumed that both the parent and product lattices are Bravais
lattices (see Definition 5) and that the transformation strain, i.e. the deformation
that maps a unit cell of the parent lattice to a unit cell of the product lattice, is
homogeneous.
The following definitions are standard and will be used throughout.
Definition 1. Let R ∈ {Z,R} denote the set of integer or real numbers respectively
and define
∗ R3×3: Vector space of matrices with entries in R.
∗ R3×3+ ∶=R3×3 ∩ {det > 0}: Orientation preserving matrices with entries in R.
∗ GL(3,R) ∶=R3×3 ∩ {A ∈R3×3 is invertible} (General linear group).
∗ GL+(3,R) ∶= GL(3,R) ∩R3×3+ : Set of orientation preserving invertible 3 × 3
matrices with entries in R.
∗ SL(3,Z) ∶= GL+(3,Z) = {A ∈ Z3×3 ∶ detA = 1} (Special linear group).
∗ SO(3) = {R ∈ R3×3 ∶ RTR = I,detR = 1} (Group of proper rotations).
∗ P24 ⊂ SO(3) (Symmetry group of a cube) - see Lemma 1 below.
Further define the multiplication of a matrix F and a set of matrices S by F.S ∶=
{FS ∶ S ∈ S}.
The following Lemma establishes a characterisation of the group P24, i.e. the set
of all rotations that map a cube to itself.
Lemma 1. Let Q = [−1,1]3 be the cube of side length 2 centred at 0 and define
P24 = {P ∈ SO(3) ∶ PQ = Q}. Then ∣P24∣ = 24 and P24 = SO(3) ∩ SL(3,Z).
Proof. Suppose that P ∈ P24 and let {ei}i=1,2,3 denote the standard basis of R3. By
linearity, P maps the face centres of Q to face centres, i.e. for each i = 1,2,3 there
exists j ∈ {1,2,3} such that Pei = ±ej . Therefore, Pki = Pei ⋅ ek = ±δkj ∈ {−1,0,1}
and thus P ∈ SL(3,Z).
Conversely, if P ∈ SO(3) ∩ SL(3,Z), its columns form an orthonormal basis and
since P has integer entries the columns have to be in the set {±ei}i=1,2,3. Hence, P
is of the form
P =3
∑j=1
±ekj ⊗ ej
and Pei = ±(ei ⋅ ej)ekj . Thus, P maps face centres of Q to face centres and, by
linearity, the cube to itself. Further, there are precisely six choices (3×2) for the
first column ±ek1 , four choices (2×2) for ±ek2 and two choices for ±ek3 . Thus, taking
into account the determinant constraint, there are 24 elements in P24.
5
2 PRELIMINARIES
Remark 1. Essentially the same proof can be used to show that SO(n)∩SL(n,Z) =
PNn , where PNn with Nn = 2n−1n! denotes the symmetry group of a n-dimensional
cube. As before the value of Nn arises from having 2×n choices for the first column
of Q, then 2× (n− 1) choices for the second column of Q, ... and the last column of
Q is already fully determined by the determinant constraint.
Definition 2. (Pseudometric and metric)
Let X be a vector space and x, y, z ∈X. A map d ∶X×X → [0,∞) is a pseudometric
if
1. d(x,x) = 0,
2. d(x, y) = d(y, x) (symmetry),
3. d(x, z) ≤ d(x, y) + d(y, z) (triangle inequality).
If in addition d is positive definite, i.e. d(x, y) = 0⇔ x = y, we call d a metric.
Definition 3. (Matrix norms)
For a given matrix A ∈ R3×3 we define the following norms
∗ Frobenius norm:
∣A∣ =√
Tr(ATA) =
¿ÁÁÀ
3
∑i,j=1
A2ij .
∗ Spectral norm:
∣A∣2 = sup∣x∣=1
∣Ax∣ =√
maxi=1,2,3
λi(ATA) = maxi=1,2,3
νi(A),
where for i = 1,2,3, νi(A) are the principal stretches/singular values of A
and λi(ATA) are the eigenvalues of ATA.
∗ Column max norm:
∥A∥2,∞ = maxi=1,2,3
∣Aei∣ = maxi=1,2,3
∣ai∣,
where {a1, a2, a3} are the columns of A.
Unless otherwise specified, here and throughout the rest of the paper ∣ ⋅ ∣ always
denotes the Euclidean norm if the argument is a vector in R3 and the Frobenius
norm if the argument is a matrix in R3×3. Additionally, we henceforth denote by
col[A] ∶= {a1, a2, a3} the columns of the matrix A and then write A = [a1, a2, a3].
The proofs of the following statements are elementary and can be found in stan-
dard textbooks on linear algebra (see e.g. [Gen07]).
Lemma 2. (Properties of matrix norms)
Both the Frobenius norm and the spectral norm are unitarily equivalent, that is
∣RAS∣ = ∣A∣ (1)
6
2 PRELIMINARIES
for any R,S ∈ SO(3). Further both norms are sub-multiplicative, that is given
A,B,C ∈ R3×3 then ∣ABC ∣ ≤ ∣A∣∣B∣∣C ∣ and thus in particular if ∣A∣∣C ∣ ≠ 0 then
∣B∣ ≥∣ABC ∣
∣A∣∣C ∣. (2)
Further the spectral norm is compatible with the Euclidean norm on R3, that is
∣Ax∣ ≤ ∣A∣2∣x∣. (3)
The following sets will be of particular importance when proving the optimality
of lattice transformations.
Definition 4. (SLk(3,Z))
For k ∈ N define
SLk(3,Z) ∶= {A ∈ SL(3,Z) ∶ ∣Amn∣ ≤ k ∀m,n ∈ {1,2,3}}
and
SL−k(3,Z) ∶= {A ∈ SL(3,Z) ∶ ∣ (A−1)mn
∣ ≤ k ∀m,n ∈ {1,2,3}}.
Clearly SLj(3,Z) ⊂ SLk(3,Z) ∀0 ≤ j ≤ k and ∣SL−k(3,Z)∣ = ∣SLk(3,Z)∣ for all k ∈ Z.
For example we have ∣SL1(3,Z)∣ = 3 480, ∣SL2(3,Z)∣ = 67 704, ∣SL3(3,Z)∣ = 640 824,
∣SL4(3,Z)∣ = 2 597 208, ∣SL5(3,Z)∣ = 10 460 024 and ∣SL6(3,Z)∣ = 28 940 280.
Below we recall some basic definitions and results from crystallography.
Definition 5. (Bravais lattice, [Bha03] Ch. 3)
Let F = [f1, f2, f3] ∈ GL+(3,R), where col[F ] = {f1, f2, f3} are the columns of
F . We define the Bravais lattice L(F ) generated by F as the lattice generated by
col[F ], i.e.
L(F ) ∶= col[F.Z×+ ].
Thus by definition a Bravais lattice is spanZ{f1, f2, f3} together with an orientation.
Definition 6. (Primitive, base-, body- and face-centred unit cells)
Let L = L(F ) be generated by F ∈ GL+(3,R). We call the parallelepiped spanned
by col[F ] with one atom at each vertex a primitive unit cell of the lattice. We call
a primitive unit cell with additional atoms in the centre of the bases a base-centred
unit cell, we call a primitive unit cell with one additional atom in the body centre
a body-centred (bc) unit cell and we call a primitive unit cell with additional atoms
in the centre of each of the faces a face-centred (fc) unit cell.
Remark 2. For any lattice generated by a base-, body- or face-centred unit cell
there is a primitive unit cell that generates the same lattice. The following table
gives the lattice vectors that generate the equivalent primitive unit cell for a given
base-centred (C), body-centred (I) or face-centred (F) unit cell spanned by the
vectors {a, b, c} ∈ R3. For our purposes all unit cells that generate the same lattice
7
2 PRELIMINARIES
primitive (P) base-centred (C) body-centred (I) face-centred (F)
{a, b, c} {a−b2, a+b
2, c} {−a+b+c
2, a−b+c
2, a+b−c
2} { b+c
2, a+c
2, a+b
2}
Table 1: Lattice vectors of a primitive unit cell that generates the same lattice.
are equivalent and in order to keep the presentation as simple as possible we will
always work with the primitive description of a lattice. However, we note that often
in the literature the unit cell is chosen such that it has maximal symmetry.
For example for a face-centred cubic lattice, the unit cell would be chosen as
face-centred and spanned by col[I] = {e1, e2, e3} so that it has the maximal P24
symmetry. However, if one considers primitive unit cells that span the same fcc
lattice, the one with maximal symmetry is given by the last entry in Table 1 and
thus spanned by col[F ], where
F =1
2
⎛⎜⎜⎜⎝
0 1 1
1 0 1
1 1 0
⎞⎟⎟⎟⎠
and has only 6-fold symmetry.
Lemma 3. (Identical lattice bases, [Bha03] Result 3.1)
Let L(F ) be generated by F = [f1, f2, f3] and L(G) be generated by G = [g1, g2, g3].
Then
L(F ) = L(G)⇔ G = Fµ⇔ gi =3
∑j=1
µjifj ,
for some µ ∈ SL(3,Z). The same result holds for a face-, base- and body-centred
unit cell.
Definition 7. (Lattice transformation)
Given two lattices L0 = L(F ) and L1 = L(G) generated by F,G ∈ GL+(3,R) we call
any matrix H ∈ GL+(3,R) such that H.L0 = L1 a lattice transformation from L0 to
L1.
Remark 3. In the terminology of Ericksen (see e.g. [PZ02] p. 62ff.), if L0 = L1 (i.e.
G = Fµ), the matrices H in Definition 7 are precisely all the orientation preserving
elements in the global symmetry group of F . Additionally, the matrices H that are
also rotations constitute the point group of the lattice. We point out that, in this
terminology, P24 is the point group of any cubic lattice.
We end this section by defining the atom density of the lattice L(F ) and relating
it to the determinant of F .
Definition 8. (Atom density)
For a given lattice L we define the atom density ρ(L) by
ρ(L) ∶= limN→∞
#{QN ∩L}
N3,
8
3 METRICS AND EQUIVALENCE ON MATRICES AND LATTICES
where QN = [0,N]3 is the cube with side-length N and # counts the number of
elements in a discrete set. Thus ρ(L) is the average number of atoms per unit
volume.
Lemma 4. Let L = L(F ) be generated by F ∈ GL+(3,R) then
ρ(L) =1
detF.
In particular a transformation H = GF −1 from L0 = L(F ) to L1 = L(G) does not
change the atom density if and only if it is volume preserving, i.e. detH = 1.
Further the atom density is well defined.
Proof. Denote by U ⊂ R3 the unit cell spanned by col[F ], so that the volume of U
is given by ∣U ∣ = detF . Taking n distinct points xi ∈ L we find that ∣⋃ni=1(xi +U)∣ =
ndetF since all elements are disjoint (up to zero measure). Let l denote the side
length of the smallest cube that contains U . Further, as in Definition 8, let QN =
[0,N]3 denote a cube of side-length N and define Q±N = [∓2l,N ± 2l]3. Then
Q−N ⊂ ⋃
x∈L∩QN(x + U) ⊂ Q+
N
and thus by taking the volumes of the sets we obtain
(N − 4l)3 ≤ #{QN ∩L}detF ≤ (N + 4l)3.
Dividing by N3 and taking the limit N →∞ yields the result. Since this limit exists
for all sequences N →∞ the density is well defined.
3 Metrics and equivalence on matrices and lattices
Definition 9. (Equivalent matrices and lattices )
We define an equivalence relation ∼ between matrices F,G in GL+(3,R) by
F ∼ G ∶⇔ ∃R ∈ SO(3) ∶ G = RF,
so that the equivalence class [F ] of F is given by [F ] = {G ∈ R3×3+ ∶ F ∼ G}. We
denote the quotient space, i.e. the space of all equivalence classes, by
GL+(3,R) ∶= {[F ] ∶ F ∈ GL+(3,R)}.
Furthermore, we define an equivalence relation ∼ between two lattices L0 and L1
by
L0 ∼ L1 ∶⇔ ∃R ∈ SO(3) ∶ L1 = R.L0.
We are now in a position to define a metric on the quotient spaces.
9
3 METRICS AND EQUIVALENCE ON MATRICES AND LATTICES
Lemma and Definition 10. (Induced metric)
Any pseudometric d ∶ GL+(3,R) ×GL+(3,R)→ [0,∞) with the property
d(F,G) = 0⇔ G = T ∗F for some T ∗ ∈ SO(3) (∗)
naturally induces a metric d on GL+(3,R) via
d([F ], [G]) = minR,S∈SO(3)
d(RF,SG).
Proof. The quantity d is clearly well defined, so that in particular d([F ], [G]) =
d(F,G) and we may henceforth drop the [⋅] in the arguments of d. We first show
that d is a metric on GL+(3,R). Positivity and symmetry are obvious from the
definition. For definiteness note that if minR,S∈SO(3) d(RF,SG) = 0 then by (∗) we
have S∗G = T ∗R∗F for some R∗, S∗, T ∗ ∈ SO(3) and thus F ∼ G. It remains to
show the triangle inequality. We have
d(F,H) = minR,S∈SO(3)
d(RF,SH)
≤ minR,S
(d(RF,G) + d(G,SH))
≤ minR,S′
(d(RF,S′G) +�����d(G,S′G)) +minR′,S
(d(SH,R′G) +�����d(G,R′G))
= d(F,G) + d(G,H),
where we have used the triangle inequality, symmetry of d and (∗).
Example 1. The family of maps dr ∶ GL+(3,R) × GL+(3,R) → [0,∞), r ∈ R/{0}
given by
dr(F,G) = ∣(FTF )r/2
− (GTG)r/2
∣
are pseudometrics such that (∗) holds. In particular, each of them induces a metric
dr on the quotient space GL+(3,R).
Proof. Positivity is obvious. The triangle inequality follows from the corresponding
property of the Frobenius norm, i.e.
dr(F,H) = ∣(FTF )r/2
− (HTH)r/2
∣
≤ ∣(FTF )r/2
− (GTG)r/2
∣ + ∣(GTG)r/2
− (HTH)r/2
∣ = dr(F,G) + dr(G,H)
and the property (∗) follows from
dr(F,G) = 0⇔ (FTF )r/2
= (GTG)r/2⇔ (FG−1
)T(FG−1
) = I
⇔ FG−1∈ SO(3)⇔ F = T ∗G for some T ∗ ∈ SO(3).
10
4 OPTIMAL LATTICE TRANSFORMATIONS
Remark 4. The metric d2 has previously been used in [Bha03, Chapter 3], where
it was defined as the distance between the metric CF = FTF = (fi ⋅ fj)ij of a set of
lattice vectors col[F ] = {f1, f2, f3} and the metric CG = GTG = (gi ⋅ gj)ij of a set of
lattice vectors col[G] = {g1, g2, g3}. The use of the term metric in [Bha03] is not to
be confused with the use of metric in the present paper.
4 Optimal lattice transformations
This section embodies the main part of the present paper. We first establish what
we mean by an optimal transformation from one lattice to another and then, for a
family of such criteria, show the existence of optimal transformations between any
two Bravais lattices.
Lemma 5. Let L0 = L(F ) and L1 = L(G) be generated by F,G ∈ GL+(3,R). Then
all possible lattice transformations from L0 to L1 are given by Hµ = GµF −1, µ ∈
SL(3,Z). In particular, the lattices coincide if and only if there exists µ ∈ SL(3,Z)
such that GµF −1 = I and they are equivalent, i.e. L0 ∼ L1, if and only if there exists
µ ∈ SL(3,Z) such that GµF −1 ∈ SO(3).
Proof. Let Hµ = GµF−1, µ ∈ SL(3,Z). Then
Hµ.L0 =Hµ.L(F ) =Hµ.col[F.Z×+ ] = col[HµF.Z×+ ]
= col[GµF −F.Z×+ ] = col[Gµ.Z×+ ] = col[G.Z×+ ] = L1,
where we used that µ is invertible over Z, so that µ.Z3×3+ = Z3×3
+ . Thus Hµ is a lattice
transformation from L0 to L1. For the reverse direction we know by Lemma 3 that
L(F ) = L(F ′) if and only if F ′ = Fµ′, µ′ ∈ SL(3,Z) and L(G) = L(G′) if and
only if G′ = Gµ′′, µ′′ ∈ SL(3,Z) so that all possible generators for L0 are given by
Fµ′, µ′ ∈ SL(3,Z) and all possible generators for L1 are given by Gµ′′, µ′′ ∈ SL(3,Z).
Thus any lattice transformation from L0 to L1 is given by
Hµ′′
µ′ = Gµ′′µ′−1F −1, µ′, µ′′ ∈ SL(3,Z).
But by the group property we may set Hµ ∶=Hµ′′
µ′ with µ = µ′′µ′−1
∈ SL(3,Z).
Definition 11. (d-optimal lattice transformations)
Given two lattices L0 = L(F ) and L1 = L(G) with F,G ∈ GL+(3,R) and a metric d,
we call a lattice transformation Hµ = GµF−1, µ ∈ SL(3,Z) (cf. Lemma 5) d-optimal
if it minimises the distance to I with respect to d, i.e. Hmin = GµminF−1, where
µmin = arg minµ∈SL(3,Z)
d(Hµ, I). (4)
If d is a pseudometric satisfying (∗), we call a transformation d-optimal if it is
optimal with respect to the induced metric d on the quotient space GL+(3,R), i.e.
11
4 OPTIMAL LATTICE TRANSFORMATIONS
the d-optimal transformation is the one that is d-closest to being a pure rotation
and maps L0 to a lattice in the equivalence class [L1] ∶= {L ∶ L ∼ L1} of L1.
Example 2. For the pseudometrics dr from Example 1 the explicit expressions for
the distance in (4) read
dr(H, I) = ∣(HTH)r/2
− I∣ = (3
∑i=1
(νri − 1)2)
1/2
, (5)
where νi, i = 1,2,3 are the principal stretches/singular values of H. The quantities
(HTH)r/2 − I are clearly measures of principal strain and are known as the Doyle-
Ericksen strain tensors (see [DE56, p. 65]). For r = 1, it is simple to verify that
d1(H, I) = dist(H,SO(3)) = minR∈SO(3)
∣H −R∣.
Remark 5. For a general metric d as in Definition 11, the optimal transformation
Hmin between L(F ) and L(G) is unchanged under actions of the point groups of
both lattices, i.e.
Hmin = G arg minµ∈SL(3,Z)
d(GµF −1, I)F −1= (PG) arg min
µ∈SL(3,Z)d((PG)µ(QF )
−1, I)(QF )−1
for any P in the point group of L(G) and Q in the point group of L(F ). Throughout
the rest of the paper, we only use pseudometrics satisfying (∗). In this case, the
notion of optimality is not only invariant under actions of the respective point groups
but also under rigid body rotations of the product lattice. Thus Definition 11 returns
an equivalence class [Hmin] ∈ GL+(3,R) of d-optimal transformations. By the polar
decomposition theorem we may, and henceforth always will, pick the symmetric
representative
Hmin ∶=
√
HTminHmin ∈ R3×3
sym,
i.e. the pure stretch component of the transformation Hmin. Note that in general
the set of minimising equivalence classes {[Hmin] ∶Hmin is d-optimal} may contain
more than one element. In such a case, different regions of the parent lattice may
transform according to any of these optimal strains, giving rise to e.g. twinning.
The pseudometrics dr are additionally invariant under rotations from the right,
i.e. dr(H, I) = dr(HS, I) for all S ∈ SO(3). For any such pseudometric, a rigid
body rotation R of the parent lattice L0 results in an optimal transformation with
stretch component RHminRT, where Hmin is the stretch component of the optimal
transformation from L0 to [L1]. Note that RHminRT is simply Hmin expressed in
a different basis and in particular, even though the coordinate representation is
different, the underlying transformation mechanism is unchanged.
Our main theorem says that a dr-optimal lattice transformation always exists
and the following lemma will be the crucial tool in the proof.
12
4 OPTIMAL LATTICE TRANSFORMATIONS
Key Lemma 6. Let H be a lattice transformation from L0 = L(F ) to L1 = L(G)
and consider a lattice vector f ∈ L0 that is transformed by H to g =Hf . Then
νmax(H) ≥ ∣g∣/∣f ∣ ≥ νmin(H),
where νmin(H), νmax(H) denote, respectively, the smallest and largest principal
stretches/singular values of H. In particular for any s > 0,
ds(H, I) = (3
∑i=1
(νsi − 1)2)
1/2
≥ maxi
∣Hfi∣s
∣fi∣s− 1, (6)
d−s(H, I) = (3
∑i=1
(ν−si − 1)2)
1/2
≥ maxi
∣H−1gi∣s
∣gi∣s− 1, (7)
where col[F ] = {f1, f2, f3} and col[G] = {g1, g2, g3}.
Proof. Consider the singular value decomposition ofH = UDV , whereD = diag(ν1, ν2, ν3)
and U,V ∈ SO(3). Then
∣g∣ = ∣UDV f ∣ = ∣DV f ∣ ≤ maxi
νi(H)∣V f ∣ = νmax(H)∣f ∣
and analogously for the lower bound.
Theorem 1. Given two lattices L0 = L(F ) and L1 = L(G) generated by F,G ∈
GL+(3,R) respectively, there exists a dr-optimal lattice transformation Hµmin=
GµminF−1, for any r ∈ R/{0}. For s > 0 all optimal changes of basis are contained
in the finite compact sets
ds ∶ {µ ∈ SL(3,Z) ∶ ∥µ∥s2,∞ ≤∥F ∥s2,∞
νsmin(G)(m0,s + 1)} , (8)
d−s ∶ {µ ∈ SL(3,Z) ∶ ∥µ−1∥s2,∞ ≤∥G∥s2,∞
νsmin(F )(m0,−s + 1)} , (9)
where νmin(A) denotes the smallest principal stretch/singular value of A and
m0,r ∶= dr(HI, I) = dr(GF −1, I).
Proof. As the minimisation is over the discrete set SL(3,Z) it suffices to show that
the minimum is attained in a (compact) finite subset of SL(3,Z) given by (8) and
(9). Let Hµ = GµF −1, µ ∈ SL(3,Z) be a lattice transformation from L0 = L(F ) to
L1 = L(G). Then, letting {ei}i=1,2,3 denote the standard basis vectors of R3×3,
Hµfi = GµF−1Fei = Gµei and H−1
µ gi = Fµ−1G−1Gei = Fµ
−1ei
13
4 OPTIMAL LATTICE TRANSFORMATIONS
and thus, by using the Key Lemma 6 and (3), we obtain
ds(Hµ, I) ≥ maxi
∣Gµei∣s
∣fi∣s− 1 ≥
∥µ∥s2,∞
∣G−1∣s2∥F ∥s2,∞− 1 =
νsmin(G)∥µ∥s2,∞
∥F ∥s2,∞− 1,
d−s(Hµ, I) ≥ maxi
∣Fµ−1ei∣s
∣gi∣s− 1≥
∥µ−1∥s2,∞
∣F −1∣s2∥G∥s2,∞− 1 =
νsmin(F )∥µ−1∥s2,∞
∥G∥s2,∞− 1,
where in the equality we have used that {νi(A−1)}i=1,2,3 = {(νi(A))−1}i=1,2,3. Thus
dr(Hµ, I) > dr(HI, I) for all µ in the complement of the respective sets given by (8)
and (9) and therefore Hµ cannot be dr-optimal.
Remark 6. The distance d1(H, I) seems to be the most natural candidate to de-
termine the transformation requiring least atomic movement. The quantities νi − 1
measure precisely the displacement along the principal axes and thus their use is
in line with the criterion of smallest principal strains as in e.g. [Bha01], [BW72]
and [Lom55]. The distance d2(H, I) seems natural from a mathematical perspec-
tive as the tensor HTH corresponds to the flat metric induced by the deformation
H and it has also been used to define the Ericksen-Pitteri neighbourhood of a lattice
(see e.g. (2.17) in [BJ92]). Finally, the distance d−2(H, I) has recently been used
in [CSTJ16] in order to avoid singular behaviour when considering sublattices.
Below we illustrate the differences of d1, d2 and d−2 through a simple but instruc-
tive 1D example.
Example 3. (A comparison of different optimality conditions)
We consider two atoms A,B that are originally at unit distance, i.e. ∣A − B∣ = 1
and then move the atom B to its deformed position B′. Thus H is simply a scalar
quantity given by H = ∣A −B′∣/∣A −B∣ = ∣A −B′∣. It can be seen from Table 2 that
B′ −B = 0.5, B′ −B = −0.5, deformation y such thatr H = ∣A −B′∣ = 1.5 H = ∣A −B′∣ = 0.5 dr(y, I) = dr(x, I)1 d1(H, I) = 0.5 d1(H, I) = 0.5 y = 2 − x
2 d2(H, I) = 1.25 d2(H, I) = 0.75 y =√
2 − x2
-2 d−2(H, I) = 0.5 d−2(H, I) = 3 y = 1√2−x−2
Table 2: Comparison of different distances
d1 depends only on the distance between B and B′; an expansion by 100% has the
same d1 distance to I as a contraction to 0, i.e. moving A onto B. The metric
d2 penalises expansions more than contractions; e.g. an expansion by ≈ 141%
has the same d2 distance to 1 as a contraction to 0. The metric d−2 penalises
contractions significantly more than expansions; e.g. an expansion by ∞ has the
same d−2 distance to I as a contraction to ≈ 70%, i.e. reducing the distance between
A and B by ≈ 30%.
14
4 OPTIMAL LATTICE TRANSFORMATIONS
A remark on the computation of the optimal transformation
Theorem 1 provides the necessary compactness result to reduce the original min-
imisation problem over the infinite set SL(3,Z) to a finite subset given by (8) and
(9) respectively. To this end, it is worth noting that the smaller the deformation
distance m0,r = dr(GF−1, I) of the initial lattice basis the smaller the radius of the
ball in SL(3,Z) that contains the optimal µ. However, in specific cases, where bet-
ter estimates are available, it might be advantageous to start with an initial lattice
basis that is not optimal.
Nevertheless, in order to explicitly determine the optimal transformations one
still needs to compare the distances dr(Hµ, I) for all elements contained in the
finite sets given by (8) and (9) respectively. This can easily be carried out with any
modern computer algebra program and possible implementations can be found in
the Appendix.
In order to ensure that the solution of this finite minimisation problem is correct
one needs to verify that the difference ∆ between the minimal and the second to
minimal deformation distance is large compared to possible rounding errors (if any).
The computations in Sections 4.1 and 4.2 for the Bain strain from fcc-to-bcc/bct
are exact and thus without rounding errors. In Section 4.3 regarding the optimal
transformation in Terephthalic Acid we find that ∆ > 0.015 which is large compared
to machine precision.
4.1 The Bain strain in fcc-to-bcc
Having established the general theory of optimal lattice transformations we apply
these results to prove the optimality of the Bain strain with respect to the three
different lattice metrics dr, r = −2,1,2, from the previous example. In these cases
we rigorously prove the optimality of the Bain strain first proposed in [Bai24].
Theorem 2. (Bain Optimality)
In a transformation from an fcc to a bcc lattice with no change in atom density,
there are three distinct equivalence classes of dr-optimal lattice transformations for
r = 1,2,−2. The stretch components are given by
Hmin ∈ {diag(2−1/3,21/6,21/6), diag(21/6,2−1/3,21/6), diag(21/6,21/6,2−1/3)},
i.e. the three Bain strains are the dr-optimal lattice transformations in a volume
preserving fcc-to-bcc transformation for r = 1,2,−2. The respective minimal metric
distances are
mmin,1 = d1(Hmin, I) =√
(2−1/3 − 1)2+ 2 (21/6 − 1)
2≈ 0.269,
mmin,2 = d2(Hmin, I) =√
(2−2/3 − 1)2+ 2 (21/3 − 1)
2≈ 0.522,
mmin,−2 = d−2(Hmin, I) =√
(22/3 − 1)2+ 2 (2−1/3 − 1)
2≈ 0.656.
15
4 OPTIMAL LATTICE TRANSFORMATIONS
Figure 2: Face-centred and body-centred cubic unit cells with equal atom density.
Proof. Let L0 denote the fcc lattice, where the fcc unit cell has unit volume and let
L1 denote the bcc lattice with the same atom density (see Fig. 2). Then L0 = L(F )
and L1 = L(B), where
F =1
2
⎛⎜⎜⎜⎝
0 1 1
1 0 1
1 1 0
⎞⎟⎟⎟⎠
= [f1, f2, f3], B = 2−1/31
2
⎛⎜⎜⎜⎝
−1 1 1
1 −1 1
1 1 −1
⎞⎟⎟⎟⎠
= [b1, b2, b3] (10)
and, in particular, detF = detB = 4−1. Let Hµ = BµF −1, µ ∈ SL(3,Z) denote the
lattice transformation from L0 to L1 (cf. Lemma 5). By definition Hµ is optimal
if µ satisfies (4). We first show the optimality with respect to d1 and d2. We
may find an upper bound on the minimum by only considering µ ∈ SL1(3,Z) (cf.
Definition 4). With the help of a computer we find exactly 72 such µ’s and all
corresponding deformations are (volume preserving) Bain strains. To complete the
proof we employ our key Lemma 6 to show that any µ ∈ SL(3,Z)/SL1(3,Z) cannot
be optimal with respect to either d1 or d2. Let µ ∈ Z3×3 be given by
µ =
⎛⎜⎜⎜⎝
α1 α2 α3
β1 β2 β3
γ1 γ2 γ3
⎞⎟⎟⎟⎠
. (11)
Then bµ,i =Hµfi = Bµei = αib1+βib2+γib3 and after dropping the index i we obtain
∣bµ∣2= (α2
+ β2+ γ2)∣b1∣
2+ 2(αβ + βγ + αγ)⟨b1, b2⟩, (12)
where we have used that ∣bi∣ = ∣bj ∣ and ⟨bi, bj⟩ = ⟨bk, bl⟩ for all i ≠ j and k ≠ l. We
compute ∣fi∣ = 2−1/2, ∣bi∣2 = 3 ⋅ 2−8/3 and ⟨bi, bj⟩ = −2−8/3. By (6) we estimate
d1(Hµ, I) ≥2−4/3 (ρ(α,β, γ))
1/2
2−1/2− 1, (13)
d2(Hµ, I) ≥2−8/3ρ(α,β, γ)
2−1− 1 (14)
16
4 OPTIMAL LATTICE TRANSFORMATIONS
where ρ(α,β, γ) ∶= α2+β2+γ2+(α−β)2+(β−γ)2+(α−γ)2. If µ ∈ SL(3,Z)/SL1(3,Z)
then ρ(α,β, γ) ≥ 8 and thus
d1(Hµ, I) ≥ 22/3 − 1 > 0.5 ≫mmin,1 and d2(Hµ, I) ≥ 24/3 − 1 > 1.5 ≫mmin,2.
To show d−2-optimality we consider Hµ = B(Fµ−1)−1 and use the ansatz (11) for
µ−1 instead of µ. We compute
∣Fµ−1ei∣2=
1
4((α + β)2 + (β + γ)2 + (α + γ)2)
and we note that bi =HµFµ−1ei. Thus by (7) we can estimate
d−2(Hµ, I) ≥14σ(α,β, γ)
3 ⋅ 2−8/3− 1 = 2 ⋅ 22/3 − 1 > 2.17 ≫mmin,−2, (15)
where σ(α,β, γ) ∶= (α+β)2+(β+γ)2+(α+γ)2 and we have used that σ(α,β, γ) ≥ 6 for
µ−1 ∈ SL(3,Z)/SL1(3,Z). Therefore the d−2-optimal µ is contained in SL−1(3,Z).
Corollary 1. The three Bain strains remain the dr-optimal lattice transformations,
r = 1,2,−2, from fcc-to-bcc if the volume changes by λ3, provided that λ > 0.84 for
r = 1, λ > 0.64 for r = 2 and λ < 1.19 for r = −2. The stretch components of the
three optimal equivalence classes are given by
Hλmin ∈ {λdiag(2−1/3,21/6,21/6), λdiag(21/6,2−1/3,21/6), λdiag(21/6,21/6,2−1/3)}.
The minimal metric distances are given by
mλmin,1 = d1(H
λmin, I) =
√
(2−1/3λ − 1)2+ 2 (21/6λ − 1)
2,
mλmin,2 = d2(H
λmin, I) =
√
(2−2/3λ2 − 1)2+ 2 (21/3λ2 − 1)
2,
mλmin,−2 = d−2(H
λmin, I) =
√
(22/3λ−2 − 1)2+ 2 (2−1/3λ−2 − 1)
2.
Proof. Replace µ→ λµ in (11) in the proof of Theorem 2. Then (13), (14) and (15)
respectively read
d1(Hλµ , I) ≥
2−4/3(λ2ρ(α,β,γ))1/2
2−1/2− 1≥ 21/6
2λ ⋅ infS ρ
1/2 − 1,
d2(Hλµ , I) ≥
2−8/3λ2ρ(α,β,γ)2−1
− 1 ≥ 2−2/3λ2 ⋅ infS ρ − 1,
d−2(Hλµ , I) ≥
14σ(α,β,γ)3⋅2−8/3λ2 − 1 ≥ 22/3
3λ2 ⋅ infS σ − 1.
If as above S = SL(3,Z)/SL1(3,Z) then infS ρ = 8 and thus d1(Hλµ , I) ≥ mλ
min,1 for
λ > 0.84 and d2(Hλµ , I) ≥ mλ
min,2 for λ > 0.64 and infS σ = 6 so that d−2(Hλµ , I) ≥
mλmin,−2 for λ < 1.19.
The following Remark concerns the relationship between the 72 minimising states.
17
4 OPTIMAL LATTICE TRANSFORMATIONS
Remark 7. (Relations between the minimal deformations for fcc-to-bcc)
Let L(F ) and L(B) be the fcc and bcc lattices respectively. Let µ0 be one of
the optimal changes of lattice basis and Hi = BµiF−1, i = 0, . . . ,71 denote the 72
optimal lattice deformations associated to the optimal changes of basis µi ∈ SL(3,Z)
given by Theorem 2. Then all optimal Hi’s and corresponding µi’s are given by
HPQ = PH0Q = BµPQF−1, P,Q ∈ P
24,
where µPQ = B−1PBµ0F−1QF ∈ SL(3,Z). We note that the latter equation holds
since P24 is the point group of both cubic lattices and thus B−1PB and F −1QF
are contained in SL(3,Z). Since there are only three equivalence classes of optimal
lattice transformations, the 72 optimal changes of lattice basis split into three sets
of 24 µPQ’s such that the 24 corresponding HPQ’s lie in the same equivalence class.
4.2 Stability of the Bain strain
Theorem 2 showed that the Bain strain is optimal in an fcc-to-bcc phase trans-
formation. In this section, we restrict our attention to r = 1,2 and show that the
Bain strain remains optimal for a range of lattice parameters in an fcc-to-bct phase
transformation. This type of transformation is found in steels with higher carbon
content. The strategy of the proof is to treat the bct phase as a perturbation of the
bcc phase. To this end, for B as in (10), let the bct lattice be generated by
BAC = diag(A,A,C)B = 2−4/3
⎛⎜⎜⎜⎜⎜⎝
−A A A
A −A A
C C −C
⎞⎟⎟⎟⎟⎟⎠
, (16)
so that C denotes the elongation (or shortening) of the bcc cell in the z-direction and
A the elongation (or shortening) in the x- and y-direction. We note that, since P24 =
SL(3,Z)∩SO(3), the lattice L(BAC) is equivalent to the lattices L(diag(A,C,A)B)
and L(diag(C,A,A)B). Further we definem0,AC
i = di(diag(21/6A,21/6A,2−1/3C), I).The following proposition provides the most important ingredient.
Proposition 1. (“The first excited state”)
In a volume preserving transformation from an fcc-to-bcc lattice the second to min-
imal deformation distances are given by
m11 ∶= min
µ∈SL(3,Z)µ≠µmin
d1(Hµ, I) ≈ 0.70 and m12 ∶= min
µ∈SL(3,Z)µ≠µmin
d2(Hµ, I) ≈ 1.64.
In particular, all Hµ with µ ∈ SL(3,Z)/SL2(3,Z) have distance strictly larger than
m1r, r = 1,2.
Proof. For brevity let us call any deformation Hµ and the corresponding change
of basis µ that has deformation distance m1r, r = 1,2 an excited state. The proof
follows along the same lines as the proof of Theorem 2. First we show with the help
18
4 OPTIMAL LATTICE TRANSFORMATIONS
of a computer that the second to minimal deformation distance within SL2(3,Z) is
given by the above and by (13) and (14) respectively applied on SL(3,Z)/SL2(3,Z)
we know that there cannot be any excited states in SL(3,Z)/SL2(3,Z).
Corollary 2. (“The first excited state” with volume change)
In a transformation from an fcc-to-bcc lattice with volume change λ3 the second to
minimal deformation distances are given by
m1,λ1 ∶= min
µ∈SL(3,Z)µ≠µmin
d1(Hλµ , I) = 2−3/2
√
25 21/3λ2 − 4 22/3 (4 +√
17)λ + 24,
m1,λ2 ∶= min
µ∈SL(3,Z)µ≠µmin
d2(Hλµ , I) = 2−3
√
305 22/3λ4 − 400 21/3λ2 + 192.
In particular all Hλµ with µ ∈ SL(3,Z)/SL2(3,Z) have distance strictly larger than
m1,λr , r = 1,2.
Theorem 3. The Bain strain is a d1- and d2-optimal lattice transformation from
fcc-to-bct with lattice parameters A,C in the range
{(A, C) ∶ C ≥ A > 0.75 and m11 − 33/2∣BAC −B∣ ≥m0,AC
1 } for r = 1,
{(A, C) ∶ C ≥ A > 0.75 and m12 − 27∣BT
ACBAC −BTB∣ ≥m0,AC
2 } for r = 2,
(cf. Figure 3). For C > A the stretch components of the optimal lattice transforma-
tions are given by HAC
min in the set
⎧⎪⎪⎪⎪⎨⎪⎪⎪⎪⎩
⎛⎜⎜⎝
21/6A 0 0
0 21/6A 0
0 0 2−1/3C
⎞⎟⎟⎠
,
⎛⎜⎜⎝
21/6A 0 0
0 2−1/3C 0
0 0 21/6A
⎞⎟⎟⎠
,
⎛⎜⎜⎝
2−1/3C 0 0
0 21/6A 0
0 0 21/6A
⎞⎟⎟⎠
⎫⎪⎪⎪⎪⎬⎪⎪⎪⎪⎭
.
The respective minimal metric distances are
mAC
min,1 = d1(HAC
min, I) = (2 (21/6A − 1)2+ (2−1/3C − 1)
2)1/2
=m0,AC
1 , (17)
mAC
min,2 = d2(HAC
min, I) = (2 ((21/6A)2− 1)
2
+ ((2−1/3C)2− 1)
2
)
1/2
=m0,AC
2 (18)
and are achieved by exactly 24 distinct µ ∈ SL(3,Z). The case C = A corresponds
to an fcc-to-bcc transformation with volume change λ3 with λ = A = C and we refer
to Corollary 1.
Example 4. A = C = 1 recovers Theorem 2. C > A corresponds to the usual fcc-
to-bct transformation for steels with higher carbon content. C =√
2A = 21/3 is the
bct lattice that is contained in the fcc lattice, i.e. d(L0,L1) = 0.
Proof (of Theorem 3). We will show that precisely 24 of the 72 µ’s that were optimal
in the fcc-to-bcc transition remain optimal. Let us start from one of those optimal
19
4 OPTIMAL LATTICE TRANSFORMATIONS
Figure 3: The range of A,C where the Bain strain remains d1-optimal (left) andd2-optimal (right).
transformations from fcc-to-bcc given by e.g.
µ0 =
⎛⎜⎜⎜⎝
1 1 1
0 1 0
0 1 1
⎞⎟⎟⎟⎠
.
We know that HAC
µ0= BACµ0F
−1 is optimal for A = C = 1, where BAC is given by
(16). The deformation distance is dr(HAC
µ0, I) =mAC
min,r with mAC
min,r given by (17) and
(18) respectively. Further there exist 24 different µ’s and corresponding HAC
µ ’s that
have the same distance. This follows as in Remark 7 with the only difference that
the point group of the bct lattice only has 8 elements. Then, due to the invariance
of d1 and d2 under multiplication from the left or right by any rotation, the 24
matrices HAC
µ trivially have the same distance to I and one may easily verify that
these HAC
µ ’s are equally split into the three equivalence classes as in the statement
of the Theorem.
The remaining 48 µ’s that were optimal for fcc-to-bcc lead to a larger deformation
distance. One of these non-optimal lattices is generated by
BAC = 2−4/3
⎛⎜⎜⎜⎜⎜⎝
−A A 0
A A 0
−C −C −2C
⎞⎟⎟⎟⎟⎟⎠
with
(d1(BACF−1, I))
2− (mAC
min,1)2= 2−5/3 (C −A) (21/6A2
+ 21/6C2− 4 + 23/2) > 0
⇐ C > A and A2+C2
>25/6
2 −√
2⇐ C > A > 0.75 (19)
20
4 OPTIMAL LATTICE TRANSFORMATIONS
and
(d2(BACF−1, I))
2− (mAC
min,2)2= 2−4/3 (C −A) (A +C) (3A2
+ 3C2− 2 22/3) > 0
⇐ C > A and A2+C2
>28/3
3⇐ C > A > 0.75 (20)
which holds true for all C > A in the range under consideration. The remaining 47
non-optimal deformations HAC
µ are all P24 related and thus have the same distance;
in particular larger than mAC
min,r.
To show the minimality of the 24 HAC
µ ’s, we make use of our result on the first
excited state to compare their distance to I against all the remaining µ’s that were
non-optimal in the fcc-to-bcc transition. In particular, we need to show that
mAC
min,i < minµ≠µmin
dr(HAC
µ , I), r = 1,2,
where HAC
µ = BACµF−1 and µmin is any of the 72 minimising µ’s from Theorem 2.
Let us set Hµ ∶=H11
µ and estimate using the properties of dr (cf. Example 1)
minµ≠µmin
dr(HAC
µ , I) ≥ minµ≠µmin
(dr(Hµ, I) − dr(HAC
µ ,Hµ)) (21)
≥m1r − max
µ≠µmin
dr(HAC
µ ,Hµ),
where m1r denotes the first excited state (cf. Proposition 1). We estimate
d1(HAC
µ ,Hµ) ≤ ∣HAC
µ −Hµ∣ ≤ ∣BAC −B∣∣µF −1∣,
d2(HAC
µ ,Hµ) ≤ ∣HAC
µTHAC
µ −HTµHµ∣ ≤ ∣BT
ACBAC −BTB∣∣µF −1
∣2
and with the help of a computer we calculate maxµ∈SL1(3,Z) ∣µF−1∣ = 33/2. Thus
a sufficient condition that the Bain strain is d1-optimal within SL1(3,Z) is that
0.75 < A ≤ C satisfy
m11 − 33/2∣BAC −B∣ ≥mAC
min,1
yielding the area drawn in Figure 3 (left). To exclude any µ ∈ SL(3,Z)/SL1(3,Z)
we replace b by bAC = diag(A,A,C)b in (12) in the proof of Theorem 2 and estimate
∣bAC∣ ≥ A∣b∣. Concluding as in (13) we arrive at
minµ∈SL(3,Z)/SL1(3,Z)
d1(HAC
µ , I) ≥ 22/3A − 1, (22)
which needs to be larger than mAC
min,1. This holds true e.g. for A ∈ [0.85,1.7]
and C ∈ [A,1.7]. The proof of the d2-optimality proceeds analogously. To obtain
d2-optimality within SL1(3,Z) we need to satisfy
m12 − 27∣BT
ACBAC −BTB∣ ≥mAC
min,2
21
4 OPTIMAL LATTICE TRANSFORMATIONS
for all 0.75 < A ≤ C which yields the area drawn in Figure 3 (right). To exclude all
elements in the complement of SL1(3,Z) we have to ensure that 24/3A − 1>mAC
min,2
which holds true e.g. for A ∈ [0.75,1.5] and C ∈ [A,1.5].
Figure 4: Extended d1-optimality (left) and d2-optimality (right) range for A andC. The dark shaded regions are obtained by a perturbation argumentabout the fixed optimal transformations Hλ0
min, λ0 = 0.9,1.1,1.3. The lightshaded regions are obtained by a perturbation argument about the (A,C)-
dependent optimal transformation Hλ(A,C)min with λ(A,C) = 0.995
√AC.
Remark 8. By employing Corollary 1 and Corollary 2 we may extend the range of
optimal parameters A and C in Theorem 3 (cf. Figure 3) by shifting the reference
point from A = C = 1 to A = C = λ. This enables us to show that the Bain strain
remains the d1- and d2-optimal lattice deformation from fcc-to-bct in a much larger
range of lattice parameters C ≥ A. The shaded regions in Figure 4 show the values
of A,C such that the deformation HAC
min remains d1- and d2-optimal.
Proof. The main idea of the proof is to consider the element Hλµ = λHµ which is
optimal in an fcc-to-bcc transformation with volume change (cf. Corollary 1) and
“closest” to HAC
µ . Thus in (21) we write
minµ≠µmin
dr(HAC
µ , I) ≥ minµ≠µmin
(dr(Hλµ , I) − dr(H
λµ ,H
AC
µ ))
and for d1 we estimate
minµ≠µmin
d1(HAC
µ , I) ≥m1,λ1 − max
µ≠µmin
∣λHµ −HAC
µ ∣ ≥m1,λ1 − λ max
µ≠µmin
∣µF −1∣∣B −BA
λCλ∣.
To obtain the d1-optimality in SL1(3,Z), we again use maxµ∈SL1(3,Z) ∣µF−1∣ = 33/2
so that m1,λ1 − λ33/2∣B − BA
λCλ∣ ≥ mAC
min,1. Condition (22) regarding d1-optimality
outside SL1(3,Z) remains unchanged. Both conditions combined yield the area
22
4 OPTIMAL LATTICE TRANSFORMATIONS
shown in Figure 4 (left). Analogously in order to show d2-optimality we estimate
minµ≠µmin
d2(HAC
µ , I) ≥m1,λ2 − max
µ≠µmin
∣λ2HTµHµ −H
AC
µTHAC
µ ∣
≥m1,λ2 − λ2 max
µ≠µmin
∣µF −1∣2∣BTB −BT
AλCλBAλCλ∣
and again use maxµ∈SL1(3,Z) ∣µF−1∣ = 33/2 to show d2-optimality within SL1(3,Z).
The condition to be d2-optimal outside SL1(3,Z) remains unchanged. Both condi-
tions combined yield the area shown in Figure 4 (right).
Remark 9. The previous estimates can be iterated, i.e. instead of picking the Hλµ
that is closest to HAC
µ one may pick any HA’C’
µ that is in the range indicated in
Figure 4 that is closest to HAC
µ .
If C ≤ A, following the proof of Theorem 3, one finds that the optimal strain
becomes
Hmin =
⎛⎜⎜⎜⎝
21/6A+C2
±21/6A−C2
0
±21/6A−C2
21/6A+C2
0
0 0 2−1/3A
⎞⎟⎟⎟⎠
+ its P24 conjugates
at least if A and C are in the regions specified in the statement of Theorem 3 with
C ≥ A replaced by A ≥ C.
4.3 Terephthalic Acid
Terephthalic Acid is a material that has two triclinic phases (Type I and II) which
are very different from each other (cf. [BJ15, p.46 ff.]). Thus any lattice transforma-
tion necessarily requires large principal stretches and, unlike in the Bain setting, it
is not clear what a good candidate for the optimal transformation would be. How-
ever, with the help of the proposed framework the dr-optimal lattice transformation
can easily be determined. The only required input parameters are the lattice pa-
rameters of the two triclinic unit cells (cf. [BB67, p.388 Table 2]) listed in Table 3.
Form a/A○ b/A○ c/A○ α/○ β/○ γ/○
I 7.730 6.443 3.749 92.75 109.15 95.95II 7.452 6.856 5.020 116.6 119.2 96.5
Table 3: Lattice parameters of the triclinic unit cells of Terephthalic Acid
To apply our analysis we first ought to convert the triclinic to the primitive
description.
Lemma 7. (Conversion from triclinic to primitive unit cell)
The triclinic unit cell with lattice parameters a, b, c and α,β, γ generates up to an
23
4 OPTIMAL LATTICE TRANSFORMATIONS
overall rotation the same lattice as
F =
⎛⎜⎜⎜⎝
a b cos (γ) c cos (β)
0 b sin (γ) c sin−1 γ (cos (α) − cos (β) cos (γ))
0 0 c (sin2 β − sin−2 γ (cos (α) − cos (β) cos (γ))2)1/2
⎞⎟⎟⎟⎠
Proof. Let col[F ] = {f1, f2, f3}. It is easy to verify that ∣f1∣ = a, ∣f2∣ = b, ∣f3∣ = c and
that ∡(f1, f2) = γ, ∡(f1, f3) = β, ∡(f2, f3) = α.
Example 5. (Primitive cells of Terephthalic Acid)
Application of the previous Lemma to the lattice parameters in Table 3 leads to
FI =
⎛⎜⎜⎜⎝
7.730 −0.668 −1.230
0 6.408 −0.309
0 0 3.528
⎞⎟⎟⎟⎠
and FII =
⎛⎜⎜⎜⎝
7.452 −0.776 −2.449
0 6.812 −2.541
0 0 3.570
⎞⎟⎟⎟⎠
(23)
(all measures in A○)
Theorem 4. (Optimal lattice transformations in Terephthalic Acid)
The unique equivalence class of d1- and d2-optimal transformations between Tereph-
thalic Acid Form I and Terephthalic Acid Form II has a stretch component given
by
Hmin =
⎛⎜⎜⎜⎝
0.820 −0.125 −0.072
−0.125 0.994 −0.146
−0.072 −0.146 1.329
⎞⎟⎟⎟⎠
(24)
with principal stretches ν1 = 0.725, ν2 = 1.033 and ν3 = 1.385. The minimal distances
are given by mmin,1 = d1(Hmin, I) = 0.474 and mmin,2 = d2(Hmin, I) = 1.035.
Proof. We apply Theorem 1 with F = FI and G = FII . We calculate ∥FI∥2,∞ =
∣FIe1∣ = 7.730 < 7.8, νmin(FII) = 3.076 > 3, m0,1 = 0.529 < 0.55 and m0,2 = 1.197 <
1.2. Further we note that if µ ∈ SL(3,Z)/SLk−1(3,Z) then ∥µ∥2,∞ ≥√k2 + 1. There-
fore by (8) any d1-optimal µ satisfies
∥µ∥2,∞ ≤∥F ∥2,∞
νmin(G)(m0,1 + 1) <
7.8
3(0.55 + 1) = 4.03 <
√42 + 1
and is thus contained in SL3(3,Z) and by (8) any d2-optimal µ satisfies
∥µ∥22,∞ ≤∥F ∥22,∞
ν2min(G)(m0,2 + 1) <
7.82
32(1.2 + 1) = 14.872
and is thus also contained in SL3(3,Z). With the help of a computer we check that
within the set SL3(3,Z) the minimum is in both cases attained at
µmin =
⎛⎜⎜⎜⎝
0 1 0
1 0 0
1 1 −1
⎞⎟⎟⎟⎠
24
5 CONCLUDING REMARKS
with corresponding Hmin =√HT
µminHµmin
as in (24), mmin,1 = d1(Hmin, I) = 0.474
and mmin,2 = d2(Hmin, I) = 1.035.
Remark 10. We have shown that the d1- and d2-optimal transformations from
Form I to Form II of Terephthalic Acid are the same. However the d−2-optimal
transformation is different and given by
Hmin,−2 =
⎛⎜⎜⎜⎝
0.852 −0.119 −0.018
−0.119 0.950 −0.197
−0.018 −0.197 1.346
⎞⎟⎟⎟⎠
with principal stretches ν1 = 0.743, ν2 = 0.977 and ν3 = 1.429. As expected the
smallest principal stretch ν1 is bigger than before since d−2 penalises contractions
significantly more than expansions. To obtain the required analytical bounds one
calculates ∥FII∥2,∞ = ∣FIIe1∣ = 7.452 < 7.46, νmin(FI) = 3.464 > 3.45 and m0,−2 =
1.080 < 1.1 to get ∥µ−1∥22,∞ < 10 (cf. (9)) and thus the optimal µ lies in SL−2(3,Z).
Remark 11. Even though [CSTJ16] also uses the distance d−2, the reported strains
are different. However, this discrepancy is not a consequence of using sublattices
but rather a consequence of using the lattice parameters from [BB67, p.388 Table 1]
for Terephthalic Acid II3 and the lattice parameters from [BB67, p.388 Table 2] for
Terephthalic Acid I. In particular, using these mixed input parameters in “OptLat”
(cf. Appendix) yields identical values.
5 Concluding remarks
The present paper provides a rigorous proof for the existence of an optimal lattice
transformation between any two given Bravais lattices with respect to a large num-
ber of optimality criteria. Furthermore, a precise algorithm and a GUI to determine
this optimal transformation is provided (cf. Appendix). As possible applications,
the optimal transformation in steels, that is the transformation from fcc-to-bcc/bct,
and in Terephthalic Acid were determined. Through Theorem 1 and with the help of
the provided algorithm/programme one is able to rigorously determine the optimal
phase transformation in any material undergoing a displacive phase transformation
from one Bravais lattice to another.
If the parent or product phases are multilattices, the proposed framework is
not a priori applicable. Nevertheless, one may still consider Bravais sublattices
of these multilattices and proceed as before. The choice of these sublattices may
come from physical consideration. However, in order to rigorously determine the
optimal transformation between two given multilattices one would need to measure
the movement of all atoms consistently, that is one would need to take into account
both the overall periodic deformation of the unit cell and the shuffle movement of
atoms within the unit cell. Establishing such a criterion would be of great interest
but lies outside the scope of the present paper.
3a = 9.54A○, b = 5.34A○, c = 5.02A○, α = 86.95○, β = 134.65○, γ = 94.8○
25
5 CONCLUDING REMARKS
Appendix
A. Mathematica
The following Mathematica code4 determines for a given k ∈ N the optimal lattice
transformation H ∶ L(F ) → L(G) within the set {Hµ = GµF−1 ∶ µ ∈ SLk(3,Z)} for
any given F,G ∈ GL+(3,R) and for any distance measure dr(H, I), r > 0. The case
r < 0 is analogous.
For the transformation from fcc-to-bcc, F and G would be given by (10) and for
the transformation from Terephthalic Acid I to II, F and G would be given by (23).
Firstly, we generate the set SLk(3,Z)(=SL):
SL = Select[Flatten[Table[{a,b,c,d,e,f,g,h,i},
{a,-k,k},{b,-k,k},{c,-k,k},{d,-k,k},{e,-k,k},{f,-k,k},
{g,-k,k},{h,-k,k},{i,-k,k}],8],Det[Partition[#,3]]==1&];
Next we generate a list(=distlist) of all values of dr(Hµ, I) for µ ∈ SLk(3,Z):
Hmu = Function[mu,G.Partition[mu,3].Inverse[F]];
distr = Function[mu,Norm[SingularValueList[Hmu[mu]]ˆr-{1,1,1}]];
distlist = distr/@SL;
Then we generate a list(=poslist) of all the positions of µ’s in SLk(3,Z) that
give rise to the minimal deformation distance:
poslist = Flatten[Position[distlist,RankedMin[distlist,1]],1];
Further we calculate the minimal deformation distance m0, the second to min-
imal deformation distance m1 and return their numerical difference ∆ = m1 −
m0(=delta):
m0 = distlist[[poslist[[1]]]]; m1 = Sort[distlist][[Length[poslist]+1]];
delta = N[m1-m0]
Finally we return a list of all µ’s that give rise to an optimal deformation Hµ and
a list of all optimal Hµ’s:
SL[[poslist]]
Hmu/@SL[[poslist]]
B. MATLAB
A Graphical User Interface (GUI) called “OptLat” can either be found on MATLAB
File Exchange5 (requires MATLAB) or downloaded directly as a standalone Windows
application6.
Acknowledgements: We would like to thank R. D. James and X. Chen for helpful
discussions. The research of A. M. leading to these results has received funding from
the European Research Council under the European Union’s Seventh Framework
Programme (FP7/2007-2013) / ERC grant agreement n○ 291053.
4The original .nb file can be found online under http://solids.maths.ox.ac.uk/programs/OptLat.nb5http://uk.mathworks.com/matlabcentral/fileexchange/55554-optlat6http://solids.maths.ox.ac.uk/programs/OptLat.exe
26
References
References
[Bai24] E.C. Bain. The nature of martensite. Transactions of the Metallurgical
Society of AIME, 70(1):25–47, 1924.
[BB67] M. Bailey and C.J. Brown. The crystal structure of Terephthalic Acid.
Acta Crystallographica, 22(3):387–391, 1967.
[Bha01] H.K.D.H. Bhadeshia. Geometry of crystals. Institute of Materials, Lon-
don, 2001.
[Bha03] K. Bhattacharya. Microstructure of martensite. OUP Oxford, 2003.
[BJ92] J.M. Ball and R.D. James. Proposed experimental tests of a theory of fine
microstructure and the two-well problem. Philosophical Transactions:
Physical Sciences and Engineering, pages 389–450, 1992.
[BJ15] J.M. Ball and R.D. James. Incompatible sets of gradients and metasta-
bility. Archive for Rational Mechanics and Analysis, 218(3):1363–1416,
2015.
[BW72] J.S. Bowles and C.M. Wayman. The Bain strain, lattice correspondences,
and deformations related to martensitic transformations. Metallurgical
Transactions, 3(5):1113–1121, 1972.
[CSTJ16] Xian Chen, Yintao Song, Nobumichi Tamura, and Richard D. James. De-
termination of the stretch tensor for structural transformations. Journal
of the Mechanics and Physics of Solids, pages –, 2016.
[DE56] T.C. Doyle and J.L. Eriksen. Non-linear elasticity. In Advances in Applied
Mechanics, volume 4, pages 53–115. Elsevier, 1956.
[Gen07] J.E. Gentle. Matrix Algebra: Theory, Computations, and Applications in
Statistics. Springer Texts in Statistics. Springer New York, 2007.
[JW48] M.A. Jaswon and J.A. Wheeler. Atomic displacements in the austenite–
martensite transformation. Acta Crystallographica, 1(4):216–224, 1948.
[KM] K. Koumatos and A. Muehlemann. A theoretical investigation of orien-
tation relationships and transformation strains in steels.
[KS30] G.V. Kurdjumov and G. Sachs. Uber den Mechanismus der Stahlhartung.
Zeitschrift fur Physik, 64:325–343, 1930.
[Lom55] W.M. Lomer. The β → α transformation in Uranium-1.4 at.% Chromium
alloy. Inst. Metals, London Monograph No. 18, pages 243–253, 1955.
[PZ02] M. Pitteri and G. Zanzotto. Continuum models for phase transitions and
twinning in crystals. CRC Press, 2002.
27